Hey guys,
I'm working on an application that uses deep learning to automatically organize images in your image folders. None of your images are sent to any cloud. I'm looking for feedback on my first demo, which can be found here:
youtube.com
Thanks.
Other urls found in this thread:
image-net.org
anyforums.com
twitter.com
>Windows
>r**dit
ebin
Interesting project do u have a github? But whats the real world application of this? Cz i was thinking that people would just properly name their file images
you cant properly name 1 million images you have downloaded on your time
it would be nice to have an AI which is able to put blonde girls on one folder and read head girls on another
No github, not OSS.
Real world application is that lots and lots of people are manually organizing their images, and this could make that easier for them. It's more about the folders than the names of the images.
>demo shirt image was easy, 'known' and duplicated
also what if an image has a person wearing a shirt holding a dog with a cat on the hood of a car lol
>No github, not OSS.
botnet dropped
It'd probably just go with something random. I plan on adding multi-tagging eventually, but most people just use folders to organize their images so I worry it'd be too much of a change.
pls no mr g
This, delete thread
Im just a novice and I'm wondering when I will be able to do stuff like this
Depends on how interested you are and how much experience you have with math / stats besides programming. You can learn pretty much everything about deep learning from the FastAI course (Jeremy Howard). That course takes just a few months in free time. He recommends at least a year of programming experience though, no other requirements.
right now
you need a dataset of 14 million images and a decent GPU.
image-net.org
here it is
get cracking faggot
/thread
Nice, I like it. How good is it at sorting gay porn?
You can actually do it with much less these days
I will never get how NEET's on Jow Forums expect people to write complex softwares and give code out for free.. baka
can you explain the whole "request a bot from the server and download it"? Do you have a server set up with the hypothetical weight that the bot will need to have? why not store that locally?
you're an absolute retard
Yeah thanks for asking. That means I should probably make it clearer in the video.
The bots contain neural networks, which it seems like you inferred already. Training a neural network takes time and resources (like a GPU). Most people don't have the right GPU's so it wouldn't work on everyone's local machines.
It's neat, I like the UI. Everything was relatively simple and you're not super transparent on methodology so to me as someone with at least a little bit of understanding it's vaguely shady. Narration was good though.
Probably throw it some curveballs like a Platypus a duck and a beaver and explain what could and what will happen. Try for things like family photos or individual portraits with lookalikes and definitively different people.
>you're not super transparent on methodology so to me as someone with at least a little bit of understanding it's vaguely shady
Thanks for this feedback, it's something I'm worried about. I feel like if I start talking about neural networks then a whole lot of people will either stop listening or start thinking its a very technical tool (it's really not). Do you have any specific questions that I could answer and make it seem less shady or murky? Thanks for the curveball ideas as well.
So you just tackle "you download bots from the server," and I'm left with the black box jist werks sort of impression, which I guess is what's kinda sketchy considering the content of the requests to the server. Seems to me like you could reverse that, assume those tags exist in the current files and run a silent upload every time it gets a nudes/drugs/evidence tag. It's probably paranoid to the extreme, but I could see a clever social engineer making due with that kind of shit.
I guess I wonder what's going on server side, is it taking a series of pre-compiled algorithms and checking it against the tags or file structure? If you have "muscle cars" and "USDM muscle" is it still going to sort and order things properly?
You are retarded. Let's use a closed source app that scans all my images but trust he dev of course it doesn't send them to the cloud goy.
cool app nate
>he's gonna steal my images
you're just as retarded as the guy afraid of his code being stolen
>App
non-free software isn't a technologically curious project people can look into, it's just a product
this is advertising, which is not allowed on Jow Forums
>people rather waste weeks upon weeks of writing a program that will do literally 2 seconds of work for them when they are saving pictures from the internet manually
Just save it in a proper folder, you fucking idiot.
The images are "scanned" by the application, but the neural networks run on your local machine, so the information isn't sent to any server. Though, your requests for bots would be sent. So the server can see the categories requested but no individual images. Not really sure how to give assurance on this, but I know it will be important to some users (I don't use Google Photos for this reason).
I disagree with this because I'm here answering questions from technologically curious people.
Okay so if I wanted a USDM Muscle category and a Muscle Car category how would it handle the overlap? Could it handle that level of specificity or are we talking first order categoricals? And say we're dealing with ultra-niche goat farming in Malaysia on the remote coast of Denenguestan, do you just have some pre-programmed things or are you scraping image results off Google to grab assets to assemble the NN?
Does it sort images by your model, meaning that it has learned your personal way of sorting and applies it, or is there a training mode in it?
Mostly scraping a few sources. So generally if your favorite image search can find plenty of results, then the bots should work well regardless of how specific it is. Regarding overlap, I plan on adding multi-tagging eventually. In the meantime it might just duplicate the images or create symlinks or something. Undecided.
It creates a new model based off of other data (images not from your computer). You have to manually specify which types of content go where.
So I watched your other videos, do you really think there's a market for a meme sorter? Are you looking to be incorporated into a larger project like an adobe add-on system? I personally can't see actually paying for this, but I can see the utility in it.
Are you planning on something like a smart sort feature to build a file system based on metadata and tags for movies/music, maybe even integrate something like Picard so it can clean tags and build a system for larger libraries using a discogs/last.fm/other scraper? Likewise with PDF files? I could see it selling to hoarders if it could do those types of operations well.
>the cloud goy
You don't use a whitelist firewall??
I've always wanted to make an image sorter. My method would involve just uploading the image to something like google and use their image suggestion return string. Maybe convert the image to a thumbnail first to save bandwidth.
dont f×Ħlder sort, tags and metadata are a lot easier to work with.